The Two Stages Subject 2's Inference Chain Doesn't Have

An engineer familiar with text-to-image-diffusion-models's inference chain (prompt safety -> prompt enhancement -> generation -> harm detection -> super-resolution cascade) proposes reusing that exact five-stage chain unmodified for the text-to-video product, arguing "the video model is just a diffusion model like the image one, so the same pipeline should work."

1. Identify the two stages this subject adds to that chain, and explain why each one is required rather than optional for a text-to-video system built on LDM.
2. Where does each new stage sit in the ordered chain, and why does that placement matter?
3. Is the engineer's "it's just a diffusion model" framing wrong, right, or partially right? Justify your answer precisely.